Thoughts, stories and ideas — in English
49 posts
1. Why do we need AI models that understand video? As platforms like OTT, short-form, and live streaming continue to grow, the volume of video being produced is exploding. This has created three real-world challenges in the field: increased time for platform-specific editing and repurposing, the burden of meta-tagging for search and classification, and growing time spent analyzing video content.
1. Why does the 'noisy neighbor' problem inevitably occur? The core of Kubernetes operations is Resource Sharing. But sharing inevitably leads to Contention. - The causal chain: 1. Multiple pods are placed on a single physical node (Bin Packing) to reduce costs. 2. A specific pod (A) experiences a traffic surge or an infinite loop in its logic. 3. Pod (A) occupies the node's shared resources (CPU, Memory, etc.) to their limits. 4. Other pods (B) on the same node suffer from resource starvation, leading to performance degradation or outages.
1. Why do we need controllers? Kubernetes is a declarative system. Users simply declare the desired state — "spin up 3 nginx instances" — without specifying how to do it. The API Server stores this Desired State in etcd...
--- Concept Map - Core Keyword Connections --- 1. What is an Infrastructure Container (pause container)? Definition - A hidden container automatically created for each Pod - Performs only infinite waiting via the pause() system call, with no app logic - The owner (anchor) of the Linux Namespace shared by containers within the Pod...
Stateless and Stateful in Kubernetes - how the reconciliation loop, Deployments, and StatefulSets work at their core.
--- I debated an entire book with AI. John Assaraf's *Having It All* — a book about the Law of Attraction. After finishing it, I was left with a question: "Is this law actually valid?" In the past, I would have looked up reviews or watched a YouTube breakdown. This time, I asked AI to play three roles — a physicist, a psychologist, and a philosopher — while I took the pro side in a round-based debate. All three attacked my arguments from their own perspectives. I defended, got attacked again. By the end, my original thinking had transformed beyond recognition. It was sharper. More honest. This post is about the methodology that emerged from that process.
I read John Assaraf's book 'Having It All.' A self-help book by the author who appeared in 'The Secret.' The Law of Attraction. Positive thoughts create a positive reality. Honestly, at first I thought it was that kind of book. The universe is on your side, wish for something and it will come true. That sort of sweet talk...
1. A Docker image is a stack of layers. What is a layer? One Dockerfile instruction = one layer. Each layer stores only the diff from the previous layer. Similar in concept to a Git commit. Layer size There is no upper limit on layer size. If 500MB worth of packages are installed via RUN yum install something, that entire 500MB becomes a single layer.
AS-IS → TO-BE AS-IS: Existing hook () > ❌ Problem: The command is directly exposed in , allowing Claude to bypass without it --- TO-BE: Updated hook + local skill 1. Hook message update 2. — Adding Phase 4 > ✅ Result: No bypass method in the hook message...
1. VM vs Container: Differences in Isolation Approach | Category | VM (Virtual Machine) | Container | | --------- | ---------------------- | ----------------------...
I came across an interview with Chamath Palihapitiya, former VP at Facebook. The insights were striking enough that I wanted to organize and share them here. His principles are as follows. These ten principles overlap in meaning at points — learn constantly, live process-oriented, stay out of debt. They all connect back to a single thread.
A resume should always be something you're continuously preparing. If you try to write it all at once later, it's hard to remember what you did, and if you don't organize the details by keyword, it becomes difficult for both the writer and the reader to understand what was actually accomplished. That's why I think a resume should be managed not as a 'final product' but as a 'running record'...
To simulate an air-gapped (closed network) environment in Minikube, the following setup was configured: 1. Start Minikube (Calico CNI) 2. Install calicoctl (version matching) 3. Set environment variables 4. Create and apply GlobalNetworkPolicy 5. Testing 6. Policy management, and more...
A manual for managing Java versions using jEnv. 1. Overview — jEnv is a CLI tool that lets you install multiple Java versions on your system and switch between them globally, per-directory (local), or per-shell session. On macOS, it automatically recognizes JDKs installed at the designated path...
> Purpose: A focused guide for resolving issues where kubelet fails to pull images from Harbor in a Minikube environment. Rather than explaining complex network architecture, this guide covers only the problem scenarios and their solutions. --- 1. Symptoms - Pods can successfully resolve service names like __INLINE_CODE_8__ via CoreDNS and communicate normally. - However, **kubelet (containerd/docker)** fails when attempting to pull images from the Harbor registry: - __INLINE_CODE_9__ - __INLINE_CODE_10__ - __INLINE_CODE_11__ - __INLINE_CODE_12__
The following is a practical strategy for running a separate app source repository and app manifest repository (GitOps repo). (Based on Argo CD + Jenkins/GitHub Actions + Harbor, with consideration for Gitea, internal registries, and air-gapped production networks)
1. Pipeline Utility Steps - Concept: Provides commonly used utility functions in Jenkins Pipeline - Key Features: - , → Read/write JSON files - , → Read/write YAML files - , , and other compression-related features...
--- 1. Project Overview Environment - Kubernetes: minikube - SonarQube: Community Edition (Helm Chart) - Jenkins: Existing Jenkins server - Language: Python (Flask application) - CI/...
Table of Contents 1. Prerequisites 2. SonarQube Server Setup 3. Jenkins Plugin Installation and Configuration 4. SonarQube Project Setup 5. Jenkins Pipeline Integration 6. Quality Gates Configuration 7. Integration Testing 8. Troubleshooting 9. Best Practices 1...
Jenkins 1. Realistic Operational Patterns - Early Stage: Create jobs via UI → Keep only Jenkinsfile in service repo and operate - Growth Stage: When jobs increase, introduce seed job + Shared Library - Mature Stage: Transition to JCasC + DSL + GitOps structure...
1. Incident Summary - Date: October 2, 2025 - Environment: Minikube single-node Kubernetes cluster (Docker driver-based) - Components Involved: Jenkins, Jenkins Agent, SonarQub...
--- 1. Installation MacBook (Homebrew): Check: --- 2. Basic Structure - Session: An independent workspace (can be created per project) - Window: A tab within a session - Pane: A split screen within a window In other words: Session > Window...
#SBOM #CI/CD #Concepts #DevOps Software Bill of Materials — SBOM stands for Software Bill of Materials. Literally, it's a "recipe list" of every component (package/library) that goes into a piece of software. - Just like a car has a parts list, software is built from countless open-source libraries, frameworks, and packages. SBOM records those components, their versions, and their origins in a structured format (JSON, SPDX, CycloneDX, etc.).
#settings --- 1. Terminal Enhancement Tools - Oh My Zsh - Zsh configuration framework - Rich autocomplete, themes, and plugins - Install: - Starship Prompt - Makes your shell prompt clean and modern...
#minikube #tunneling 1. Install HAProxy Ingress > This chart sets the default IngressClass name to haproxy, so you need to use ingressClassName: haproxy in your Ingress resource for it to match. HAProxy Tech...
Control Plane & Worker Runtime - etcd: cluster state store (Raft consensus) - kube-apiserver: front door for all requests (authentication/authorization/validation) - kube-controller-manager: maintains desired state of objects (ReplicaSet, Node/Job, and other controllers...
#Infra #IaaS #Concepts #Proxy 1. Common Disadvantages - Additional Latency - Since an intermediate server is inserted between the client and the server, network latency can occur during the request/response process. - Performance degradation is especially noticeable when encryption (SSL/TLS) processing or complex rules are involved...
#Infra #Concepts #On-premises 1. Let's start with what storage even is - Your computer has storage space (SSD, HDD), right? - Servers are the same — they need disks to store data. - But enterprise data centers have dozens to hundreds of servers, so managing disks separately on each one is a nightmare...
#Infra #On-premises #Concepts 1. Cisco C9500 Overview - Official name: Cisco Catalyst 9500 Series Switch - Type: Enterprise/data center core and aggregation switch - Placement: Can be used as ToR (Top of Rack) or deployed at the Core layer...
#CRI #Infra #IaaS #Concepts 1. What is a Container Runtime? - A program that runs, stops, and deletes containers - Examples: Docker Engine, containerd, CRI-O, rkt (legacy) - Kubernetes does not run containers directly; instead, the kubelet sends requests to the runtime,...
#Infra #IaaS #Concepts #LayerStructure 1. The "Layer Map" to Build in Your Head - Edge (Global Edge): CloudFront, WAF, DNS (Route53/GA). CDN · DDoS · global routing. - L4 Ingress (Transport Layer): NLB, (on-premises L4 switch)...
#Infra #IaaS #Concepts #Proxy 1. What is a Proxy Server? - An intermediary server that acts as a middleman between a client and a server. - Instead of passing requests and responses directly, the proxy server handles them on behalf of either party. - Used for security, performance optimization, access control, load balancing, and more.
#GitOps #Infra #DevOps #Concepts 1. Definition - GitOps is a methodology for managing Operations through Git - All desired states of infrastructure/applications are declaratively defined in Git - Controllers like Argo CD and Flux continuously synchronize Git with the actual environment...
#IaaS #Onboarding #Concepts 1. Basic Concepts - Kubernetes (K8s): A container orchestration tool. Automates container deployment, scaling, networking, and resource management. - Cluster: The overall system in which Kubernetes operates....
--- 1) NodePort / LoadBalancer (Accessing via IP:Port without a hostname) Only the client-side DNS (public DNS) is used here — CoreDNS is not involved. - (1) **Public DNS** resolves to the LB IP (Route53/Cloud DNS, etc.)...
#IaaS #Onboarding #Concepts 1. HPA (Horizontal Pod Autoscaler) Issue Colleague: "Why doesn't our service automatically scale up Pods even when CPU is maxed out?" Chris: "HPA relies on the Metrics Server to pull CPU/memory usage. If the Metrics Server isn't installed or lacks RBAC permissions, the values come back as 0 and HPA won't trigger. Also, if there are no Resource Requests set on the Deployment, HPA can't calculate a baseline threshold. So you need to verify the Metrics Server is working properly and check that Requests are configured."
#IaaS #Onboarding #Concepts #Octavia --- 1) Core Load Balancing Concepts - Goal: Distribute traffic evenly across multiple backends (servers/pods) to ensure availability and performance. - L4 vs L7 - L4: TCP/UDP-level distribution. Fast and simple. Strong for source IP preservation, static IPs, and pass-through use cases. - L7: HTTP/HTTPS-level distribution. Provides application-aware features like path/host-based routing, header inspection, and cookie stickiness.
#Infra #IaaS #Concepts #Monitoring 1. Horizon - Refers to OpenStack Horizon. - The dashboard (Web UI) of OpenStack (open-source cloud platform). - Roles: - Create/delete VMs - Network management - Storage management...
#Infra #IaaS #Concepts | Term | General Infrastructure Perspective | On-Premises (OpenStack/KVM) | Cloud (AWS/EKS/ECS)...
Pod Outbound Traffic Flow 1. DNS Query (CoreDNS → External DNS) - Application inside the pod makes a request → DNS query to CoreDNS - CoreDNS forwards to an upstream DNS specified in (e.g., Google 8.8.8.8, cloud-provided DNS) → Obtains IP and responds to the pod...
#Infra #IaaS #kube #concept #rolling 1. Rolling Update A zero-downtime deployment strategy that replaces Pods sequentially with a new version. Example: replacing image version v1 → v2 --- 2. Rollout Commands for tracking and managing the deployment process and its history...
#Infra #On-premises #Concepts 1. What is SAN? - SAN (Storage Area Network) = a dedicated storage network - Connects servers (compute equipment) and storage (disk equipment) over a separate network, allowing servers to use storage as if it were a local disk...
1. Starting with Pod — in Kubernetes, a Pod = the unit of execution. - It's like a "mini server" that bundles one or more containers (like Docker) and runs them together. - Most Pods contain just one container → running a single app. - But in some cases, you can put multiple containers inside a single Pod.
#IaaS #Onboarding #Concepts #Cinder 1. Basic Storage Types Storage can be broadly categorized into three types: 1. Block Storage - Data is stored in "block units." - From the OS perspective, it looks like a local disk....
#IaC #kube #concepts 1. Basic Unit - Pod - The smallest deployable unit in Kubernetes - One or more containers + shared storage (Volumes) + network (IP) + execution info - Containers within the same Pod share IP and port space...
#Infra #On-premises #Concept 1. What is a Rack? Data centers have large metal shelf-like frames used to stack servers and network equipment. That's called a Rack. - A single rack holds multiple servers, storage units, and network devices lined up - Servers like the Dell R760 and R730xd you've seen are all rack-mounted equipment...
#Infra #IaaS #Concepts #LayeredArchitecture 1. What is a Hypervisor? - Definition: Virtualization software that allows multiple Virtual Machines (VMs) to run on a single physical server (Host). - In other words, it acts as an intermediary manager that slices up physical resources such as CPU, memory, network, and storage so that multiple VMs can share them...
#IaaS #Onboarding #Concepts #Cilium 1. Networking Basics - When you have multiple servers/applications, they need IP addresses and ports to communicate with each other. - Data flows through transport layers like TCP/UDP, and the actual requests/responses we use happen at the application layer like HTTP...
#Infra #Proxy #Concepts HAProxy (High Availability Proxy) is, as the name suggests, an open-source reverse proxy + load balancer software that emphasizes High Availability. In other words, it's one of the representative solutions designed to overcome the limitations of proxies (e.g., single point of failure, performance degradation) discussed earlier.